A dependable long-term hydrologic prediction is essential to planning, designing and management activities of water resources. A three-stage indirect multi-step-ahead prediction model, which combines dynamic spline interpolation into multilayer adaptive time-delay neural network (ATNN), is proposed in this study for the long term hydrologic prediction. In the first two stages, a group of spline interpolation and dynamic extraction units are utilized to amplify the effect of observations in order to decrease the errors accumulation and propagation caused by the previous prediction. In the last step, variable time delays and weights are dynamically regulated by ATNN and the output of ATNN can be obtained as a multi-step-ahead prediction. We use two examples to illustrate the effectiveness of the proposed model. One example is the sunspots time series that is a well-known nonlinear and non-Gaussian benchmark time series and is often used to evaluate the effectiveness of nonlinear models. Another example is a case study of a long-term hydrologic prediction which uses the monthly discharges data from the Manwan Hydropower Plant in Yunnan Province of China. Application results show that the proposed method is feasible and effective.
- Adaptive time-delay neural network
- Indirect multi-step-ahead prediction
- Spline interpolation
- Time-delay neural network
ASJC Scopus subject areas
- Water Science and Technology